Lavinia
Gambelli†
ab,
Michail N.
Isupov
c and
Bertram
Daum
*bd
aCollege of Engineering, Mathematics and Physical Sciences, University of Exeter, Exeter, EX4 4QF, UK
bLiving Systems Institute, University of Exeter, Exeter, EX4 4QD, UK. E-mail: b.daum2@exeter.ac.uk
cHenry Wellcome Building for Biocatalysis, Biosciences, College of Life and Environmental Sciences, University of Exeter, Exeter, EX4 4QD, UK
dCollege of Life and Environmental Sciences, University of Exeter, Exeter, EX4 4QD, UK
First published on 5th August 2022
Helical reconstruction is the method of choice for obtaining 3D structures of filaments from electron cryo-microscopy (cryoEM) projections. This approach relies on applying helical symmetry parameters deduced from Fourier–Bessel or real space analysis, such as sub-tomogram averaging. While helical reconstruction continues to provide invaluable structural insights into filaments, its inherent dependence on imposing a pre-defined helical symmetry can also introduce bias. The applied helical symmetry produces structures that are infinitely straight along the filament’s axis and can average out biologically important heterogeneities. Here, we describe a simple workflow aimed at overcoming these drawbacks in order to provide truer representations of filamentous structures.
Electron cryo-microscopy (cryoEM) has enabled the structural determination of a plethora of protein filaments through an image processing approach called helical reconstruction. For this, a sample of purified filaments is vitrified on electron microscopy grids and usually hundreds to thousands of micrographs are then recorded in an electron microscope. Since most biological filaments have intrinsic helical symmetry, each subunit within a filament adopts a different but defined orientation with respect to the image plane. This means that when the orientation and position of each filament subunit with respect to the filament’s axis is established, it becomes possible to calculate a 3-dimensional reconstruction.6
Classically, this approach involved the use of Fourier–Bessel analysis, whereby the diffraction pattern of a filament calculated by a Fourier transform of the image is used to determine the screw operations, or helical parameters. These helical parameters define the relative displacement of the symmetry-related subunits within the helix and are known as helical rise (the translational offset between adjacent asymmetrical units), twist (the angular offset between adjacent asymmetrical units), pitch (the length of a complete helix turn) and number of subunits per turn (NUT). If either rise and twist or pitch and NUT can be determined, the three-dimensional structure of the filament can be reconstructed in a process known as Fourier–Bessel inversion.6,7 However, the classical Fourier–Bessel method relies on high-quality diffraction patterns that are only obtained from highly ordered and relatively straight filaments – features that rarely apply to often undulating biological specimens. Moreover, the determination of the screw operations from diffraction patterns is sometimes precluded by the overlap of Bessel helical layer lines.8,9
More recently, single particle approaches have been developed, which do not depend on high quality diffraction of the raw data. This strategy was first implemented with IHRSR,9 later in SPRING10 and more recently in the Relion11 and cryoSPARC12 software packages. In a nutshell, the filaments are first divided up into segments, which are then initially treated as single particles, meaning that they are aligned with a reference and averaged to yield a projection of the reference with greatly enhanced signal to noise ratio. As in the single particle approach, the data can be classified to 2D classes to account for heterogeneity in the sample.13 To obtain the screw operators that are important for the 3D reconstruction, the 2D class averages are analysed both in real and Fourier space. For filaments of low order, rise and pitch values can sometimes be estimated from the projection. However, analyzing the layer lines of the diffraction pattern in Fourier space usually gives more precise clues and is the only source for estimating the twist value from the projection data.9 By applying the screw operators, the data are then iteratively aligned with a reference and back-projected into a 3D map. In this process, the helical parameters can be interactively refined until the reconstruction reaches its final resolution (does not improve with additional iterations).9
Unfortunately, layer line profiles can often be inconclusive (e.g. through Bessel overlap or appear to suggest more than one solution to the helical parameters).8,9 In those cases, the layer line profiles are used to narrow down sets of helical parameters that then need to be tested. This is aided by helical symmetry search routines implemented in software such as SPRING11 and cryoSPARC.12 During this process, all sets of helical parameters that do not produce a biologically meaningful 3D reconstruction are discarded and the set that does is considered as correct. For example, at a Gold-Standard resolution under ∼4.6 Å, secondary structure features (α-helices, β-strands and loops) should be resolved and α-helices should be right handed. These maps are then taken forward to build an atomic model and further interpretation.
Using the structure of an archaellum as an example, we highlight how this approach can be a trap, as more complex symmetries can be disguised by seemingly correct helical parameters. We further suggest a way out of this trap, whichcould be useful as a default step for any helical reconstruction project.
Fig. 1 Helical reconstruction of archaella from M. villosus. (a) Examples of 2D classification of the M. villosus archaellum obtained in Relion 3.1. (b) Layer line profiles of the M. villosus (Mvi, top) and M. hungatei (Mhu, bottom [re-used from ref. 17 with permission]) filaments. (c) Examples of 3D classifications obtained in Relion 3.1. Classes 1 and 4, which show typical features of archaella (central α-helix bundles (blue) and peripheral globular domains (mauve)), were selected for further processing. Scale bar, 50 Å. |
Archaellum operons usually encode one to seven archaellum subunits called archaellins15,16 and based on previously published structures, it is believed that only one of these archaellins assembles into the bulk of the archaellum filament.4,17,18M. villosus is no exception, with 3 archaellins called ArlB1, 2 and 3 encoded in the operon. With this in mind, we modelled the archaellin ArlB2 into our map, which appeared to fit the protomer densities best (Fig. 2e and f). However, the outer regions of the filament were less well resolved, making the backbone harder to trace than in the core.14
Curious as to whether we could resolve different conformations of the inherently flexible filament, we imported the refined particles into CryoSPARC 3.1.0 to perform 3D variability analysis (3DVA). As the first step of this analysis, a new refinement was performed using the helical refinement (BETA) algorithm without imposing helical parameters. Strikingly, this resulted in a reconstruction with 3.28 Å resolution, where the filament appeared to consist not of one, but two distinct subunits that alternated throughout the filament (Fig. 2g–i, Fig. 3a and b). This was confirmed by a round of 3D refinement with no helical symmetry in Relion 3.1.14
Through careful Real Space analysis using a net diagram (Fig. 3c), we found that the originally imposed helical parameters of 5.57 Å rise and 108° twist did not describe the filament’s alternating architecture correctly. Curiously, the filament could not be described as one having n + 2 helical symmetry by doubling of its helical parameters to 11.14 Å rise and 216° (−144°) twist. Instead, the minimal transformation of the M. villosus archaellum in which each monomer superimposes onto its equivalent was n + 6, resulting in revised helical parameters of 33.4 Å rise and −71.8° twist.13 Employing these helical parameters in a final round of refinement resulted in a map with improved resolution of 3.08 Å (Fig. 3d and e). The map showed a significantly improved quality of the outer regions compared to the previous one reconstructed with the erroneous helical parameters (Fig. 2e and f) and also compared to the map obtained by helical relaxation (Fig. 3e and f). The densities for glycan moieties that decorate the archaellum were better defined (asterisks in Fig. 2j) and we could unambiguously model the sequences of ArlB1 and ArlB2 into the new map, resulting in the structures shown in Fig. 2j and k.14
In the complicated symmetry of the structure, ArlB1 and ArlB2 alternate along the three left-handed 3-start helices that make up the archaellum. Interestingly, every third 3-start helix is out of register with respect to the previous two. The biological significance of this register shift is so far not clear. However, it can be assumed that this shift causes a screw axis asymmetry that may bias the superstructure of the filament into the stable superhelix that is observed in gyrating archaella.14
As exemplified by our data, it is advisable to perform symmetry relaxation even in cases where the structure appears to be biologically meaningful and shows the structural details that are expected at the resolution. This strategy will potentially uncover unexpected heterogeneities that are averaged out by the imposed symmetry and in addition allow the investigation of filament flexibility, as implemented in CryoSPARC’s 3DVA.12 Furthermore, we suggest that the maps resulting from unsymmetrised 3D refinements should be included in the validation process. This could entail a Fourier shell correlation between the symmetrised and unsymmetrised maps (Fig. 3f), as well including the unsymmetrised maps in data deposition to the Unified Data Resource for 3-dimensional Electron Microscopy (EMD).
Footnote |
† MRC Laboratory of Molecular Biology, Cambridge Biomedical Campus, Francis Crick Ave, Trumpington, Cambridge CB2 0QH. |
This journal is © The Royal Society of Chemistry 2022 |